By Nuru Shaba
AI-generated Image: The host period is too short.
Ken Griffin recently told a room of investors that AI now performs in hours what teams of PhDs used to take months to do. Six months earlier the comparison was weeks. A year before that it was a punchline.
That kind of compression is not a productivity story. It is an epidemiology problem.
Viruses have an R number. It tells you how many people one infected person is likely to infect. Below 1, the virus politely disappears. Above 1, it starts making weekend plans. Much higher and governments close airports, cancel weddings, and everyone becomes an expert in exponential curves.
Epidemiologists decompose R₀ into three components: β × c × D. The probability an exposure converts to infection (β), the rate of contact between hosts (c), and the duration each host remains infectious (D). The trick of a successful virus is not raw speed. It needs a host period long enough to travel. Too weak, and it dies out. Too strong, and it terrifies everyone into stopping movement. The most successful pathogens spread just slowly enough to be ignored.
Seasonal flu sits at roughly 1.3. Smallpox around 6. Measles around 15. COVID’s Omicron variant arrived at about 9 and we shut the world.
Now apply the same maths to a technology. R_tech = β × c × D, where β is the probability a person who tries it keeps using it, c is the number of others each user exposes to it, and D is how long the change stays invisible enough to be tolerated.
The telephone took 75 years to reach 50 million users. Facebook took about four years to reach 100 million. ChatGPT got there in two months. Whatever β, c, and D are doing in the AI case, they are doing it at a speed that pushes R past Omicron and into measles territory.
Human culture likes change, but only when it arrives wearing sensible shoes. We enjoy progress after it has filled in the correct forms, met the neighbours, and agreed not to affect house prices.
The washing machine was acceptable because it did not email the accountant. The smartphone was tolerated because it first pretended to be a phone. Even the internet took years to move from “strange thing for academics” to “where your uncle now gets political opinions from a man in sunglasses.”
AI has skipped the courtship phase. It arrived as a weather system. It changes skills, jobs, power consumption, investment flows, education, software, law, art, fraud, and customer service before most institutions have finished forming a committee to investigate the implications of digital transformation.
That is not change. That is something happening to you.
The danger for AI is not that it is useless. The danger is that it is too useful, too quickly, in too many places, for too many people to remain calm.
Slow technologies get to be called infrastructure. Fast ones get called to testify.
If AI had improved gently over twenty years we would call it productivity software and invite it to conferences. Instead it has gone from party trick to junior analyst to strategic adviser while half the world was still trying to remember its ChatGPT password.
That speed creates resistance. Governments will resist it because it threatens control. Regulators will resist it because regulation is what institutions do when they see something moving faster than their filing system. Incumbents will resist it because they are large, profitable, and allergic to surprises.
And once resistance starts, it may not be elegant. It may be clumsy, blunt, political, and wildly overcorrective. The kind of regulation that treats a spreadsheet, a chatbot, and a rogue superintelligence as the same animal because they all contain the word “algorithm.”
AI’s R is astonishingly high. One user infects a team. One team infects a company. One company infects an industry. But if the host period is too short, if the disruption becomes visible too quickly, society develops antibodies. The antibodies are fear, job protection, litigation, licensing, compliance departments, and government inquiries chaired by people who still print their emails.
AI may not be stopped because it is weak. It may be stopped because it is strong too soon.
The most successful technologies do not merely solve problems. They give society time to pretend the change was its own idea. AI has not done that. It has kicked down the door, solved the homework, rewritten the job description, and asked why the electricity bill has tripled.
That may be brilliant. But evolution does not reward brilliance. It rewards organisms that did not alarm the neighbours.
About the Author
Nuru Shaba writes on markets, technology, and how systems absorb change. His previous contributions to Canary Compass include “The Safaricom Playbook: Lessons for the Insurance Industry” and “Kenya Finance Bill of 2024: The Unintended Consequences of the Motor Vehicle Tax on Consumer Behavior and the Insurance Market.” This is his third article for the publication.